-V2-
Overall improvement of the model, this is technically version 4 but i skipped a few cos they weren't really worth uploading, hence why it took so long cos I only really make them for my own use.
I'll keep this short since nobody reads this stuff anyways.
Euler A
30-50 Steps
CFG 3-5
VAE is not baked. Again I still just use the one mentioned further down.
Prompts are in the previews.
Forge still has an issue with hashes so it probably wont show the model used when uploading to civit if you use forge.
-V1-
So since illust is back to just using quality tags instead of ponys score tags heres the tags I use. Obviously you don't have to use them they're just what give me nice results.
masterpiece, best quality, amazing quality, very aesthetic, absurd, newest, detailed eyes,
When it comes to styles like flat anime and things, for flat anime add anime to positive and 3d, realistic to negative. You can add flat color to positive too if you want a more flat style. Essentially tag what you want and negative tag what you don't want style wise.
------
Info on the model,
Wasn't planning on making an illust model then started using illust and realised I like it more than pony. Trained on cherry picked images along with merging in a few loras I made and partial merge of other models.
-----------
-VAE IS NOT BAKED-
I forgot to bake the vae in before i deleted the previous versions. May add a baked version later.
I highly recommend this vae https://civarchive.com/models/152040?modelVersionId=1023774
The vae above adds a nice amount of colour without adding so much that it looks cooked.
-Recommended Settings-
Regular SDXL resolutions
I personally swap between Euler A and dpm++ 2m sde depending on if i want more detail (dpm) or more anime (euler a) but ive tested it on other schedulers and it works fine so just use what you like.
Sheduler I just use Karras
CFG 4-7. I usually stick to 4 and have 7 on hi-res fix CFG. Just to give it more freedom but still keeping the same sharpness.
Steps: 36, I havn't changed it from 36 so again thats just my preference.
I use Adetailer for faces unless the face is close, when the face is close to the screen Adetailer is usually a quality hit depending on your resolution.
Recommend Hires Fix for higher quality images but thats just standard now. 0.3-0.4 denoise and 10-15 steps. I use 4x_NMKD-YandereNeoXL_200k upscaler since it doesn't flatten the colours like other anime upscalers.
Description
Vae isnt baked
FAQ
Comments (2)
Your model must have some comments and more attention. That's the problem with nowadays media/buzz with rating systems, if you're not well knows it's truly hard to get noticed. Pretty unfair. To me as a casual user this model is one of the best, I'm glad I found it! Haven't tried many illustrious variants, something like novaAnime or stein is on par, but I still prefer this model. You don't seem publish many models, I can only assume you prefer to reveal only work that is worth it. So, please, if you get some interesting merge that is different from the current model and no less excellent, do publish it, would be great to see more of your work. Cheers!
Thank you, I really appreciate it. When it comes to posting models I only post the ones I personally would use since I don't really like having to switch models to do different things so I try to make one that can do almost all of it. I do a lot of experimental merges and stuff but most of the time they don't turn out at what I would say was an acceptable quality to post so I just delete them usually. If I wouldn't use it then in my opinion it isn't really good enough for me to post.
If I get anything I think is good or interesting though I will for sure post it.
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.